For vertebrate organisms where a reference genome is not available, de novo transcriptome assembly enables a cost effective insight into the identification of tissue specific or differentially expressed genes and variation of the coding part of the genome. However, since there are a number of different tools and parameters that can be used to reconstruct transcripts, it is difficult to determine an optimal method. Here we suggest a pipeline based on (1) assessing the performance of three different assembly tools (2) using both single and multiple k-mer (MK) approaches (3) examining the influence of the number of reads used in the assembly (4) merging assemblies from different tools. We use an example dataset from the vertebrate Anas platyrhynchos domestica (Pekin duck). We find that taking a subset of data enables a robust assembly to be produced by multiple methods without the need for very high memory capacity. The use of reads mapped back to transcripts (RMBT) and CEGMA (Core Eukaryotic Genes Mapping Approach) provides useful metrics to determine the completeness of assembly obtained. For this dataset the use of MK in the assembly generated a more complete assembly as measured by greater number of RMBT and CEGMA score. Merged single k-mer assemblies are generally smaller but consist of longer transcripts, suggesting an assembly consisting of fewer fragmented transcripts. We suggest that the use of a subset of reads during assembly allows the relatively rapid investigation of assembly characteristics and can guide the user to the most appropriate transcriptome for particular downstream use. Transcriptomes generated by the compared assembly methods and the final merged assembly are freely available for download at http://dx.doi.org/10.6084/m9.figshare.1032613.